HybridSeg is a novel framework for tweet segmentation that splits tweets into meaningful segments to preserve semantic and context information for downstream applications like named entity recognition. It finds the optimal segmentation of a tweet by maximizing the sum of "stickiness scores" of segments, which considers global context from sources like n-grams and local context from linguistic features within a batch of tweets. Experiments show HybridSeg significantly improves segmentation quality over using just global context. It also achieves high accuracy on named entity recognition tasks.